Web Application and AI for Systemic Agronomy Design

This was research aimed at establishing a knowledge representation model and developing a computer tool to visualize and explore this knowledge. This work took place between April 2018 and December 2020, within the framework of Axis 2 of the European CASDAR AgroEcopérenne and Tradévi projects. They took place successively within the MISTEA and SYSTEM research units, located on the La Gaillarde campus, then with the French Institute of the Vine (IFV) in Grenoble.

The main challenges were to design a computer tool facilitating design in systemic agriculture, thus enabling a common representation of knowledge for the diagnosis and design of viticultural and cider systems (perennial plants). This tool was also to allow knowledge exploration, including the application of artificial intelligences to discover new interactions, particularly to understand the causes of vine decline in France.

The involved entities included the mixed research units of INRAE such as SYSTEM (now ABSys) and MISTEA (Mathematics, Informatics and STatistics for Environment and Agronomy), the Institute for Research in Horticulture and Seeds (IRHS) and the Integrated Experimental Research Unit (UERI), as well as the French Institute of Vine and Wine (IFV) and the French Institute of Cider Productions (IFPC).

Tasks & Objectives

My role within this project was that of developer, designer, and IT administrator, responsible for creating and managing the entire technology suite. As an expert in artificial intelligence and web semantics, I had to propose a model allowing the pooling of knowledge from multiple disciplines and specialties at different scales.

The main objective was to develop a computer tool to acquire, represent, and explore this knowledge. The success criteria included defining a common language for the various specialties related to agriculture and building a tool facilitating knowledge acquisition and exploration.

Actions and Development

To achieve these objectives, I co-constructed a knowledge representation model, allowing the connection of different scales and domains of agronomy. I also developed a highly configurable web application for manipulating graphs, creating two software configurations for knowledge restitution and exploration, while respecting the established model. I animated numerous workshops around this theme and co-supervised a fixed-term contract employee responsible for using the tool to help less technical users appropriate this knowledge.

The steps began with understanding the issues, revealing that knowledge restitution required a web tool for creating structured graphs. After sufficient aggregation of knowledge in the database, I created a tool allowing efficient exploration. This process was punctuated by meetings and workshops with agronomists and other stakeholders.

I collaborated with the IFV project manager, the UMR SYSTEM scientific manager, and the agronomy study engineer, establishing numerous interactions during workshops with the different engaged institutes.

Among the challenges encountered, we note the difficulty of building a model covering all domains and scales, the need to connect knowledge to existing data (related to linked data and semantic concepts), the construction of a complex IT stack (including frontend, backend, and integration pipelines), and an effort to abstract the application layer to ensure rapid and efficient evolution of the model.

Technical decisions were oriented towards the needs of the application and the domain, which led to the choice of a graph-oriented database, high-level modeling to encompass various domains and scales, and an administration interface allowing data moderation.

Results

The results led to the creation of a tool called Sygnal, which aggregated more than 2000 nodes connected by more than 24000 relationships. This tool was tested by post-doctoral students during a workshop, and the feedback was very positive, highlighting its ergonomics and interest for knowledge exploration. A scientific article was also published, bringing additional visibility to the project.

This experience, the longest of my career, allowed me to develop a complete software product from a concrete problem. I considerably improved my skills in interdisciplinary communication, and the challenges related to building and managing a graph tool for data representation and acquisition broadened my IT perspective, particularly regarding semantics and ontologies.

Technical Stack

The project relies on the following tools and technologies:

  • Frontend : TypeScript, Angular 2, HTML/CSS
  • Backend : Python
  • Database : RDF/SPARQL, Neo4j/Cypher
  • Infrastructure : Linux, servers

It is important to note that this stack was chosen for its coherence and ease of use. The major technical challenges encountered include:

  • Difficulties when restituting new knowledge in the form of graphs, particularly for establishing relationships between elements already existing and those not yet in the database
  • Notions of "local" identifier and "global" identifier, as well as automatic reconciliation of these two concepts when possible